Venue
- ICLR 2023
Date
- 2023
Twofer: Tackling Continual Domain Shift with Simultaneous Domain Generalization and Adaptation
Chenxi Liu*
Lixu Wang
Chen Sun*
Xiao Wang*
Qi Zhu*
* External authors
ICLR 2023
2023
Abstract
In real-world applications, deep learning models often run in non-stationary environments where the target data distribution continually shifts over time. There have been numerous domain adaptation (DA) methods in both online and offline modes to improve cross-domain adaptation ability. However, these DA methods typically only provide good performance after a long period of adaptation and perform poorly on new domains before and during adaptation, especially when domain shifts happen suddenly and momentarily. On the other hand, domain generalization (DG) methods have been proposed to improve the model generalization ability on unadapted domains. However, existing DG works are ineffective for continually changing domains due to severe catastrophic forgetting of learned knowledge. To overcome these limitations of DA or DG in tackling continual domain shifts, we propose Twofer, a framework that simultaneously achieves target domain generalization (TDG), target domain adaptation (TDA), and forgetting alleviation (FA). Twofer includes a training-free data augmentation module to prepare data for TDG, a novel pseudo-labeling mechanism to provide reliable supervision for TDA, and a prototype contrastive alignment algorithm to align different domains for achieving TDG, TDA, and FA. Extensive experiments on Digits, PACS, and Domain Net datasets demonstrate that Twofer substantially outperforms state-of-the-art works in Continual DA, Source-Free DA, Test-Time/Online DA, Single DG, Multiple DG, and Unified DA&DG. We envision this work as a significant milestone in tackling continual data domain shifts, with improved performance across target domain generalization, adaptation, and forgetting alleviation abilities.
Related Publications
Existing collaborative self-supervised learning (SSL) schemes are not suitable for cross-client applications because of their expensive computation and large local data requirements. To address these issues, we propose MocoSFL, a collaborative SSL framework based on Split Fe…
Knowledge Distillation (KD) is a typical method for training a lightweight student model with the help of a well-trained teacher model. However, most KD methods require access to either the teacher's training data or model parameter, which is unrealistic. To tackle this prob…
Continual Test-time Adaptation (CTA) is a promising art to secure accuracy gains in continually-changing environments. The state-of-the-art adaptations improve out-of-distribution model accuracy via computation-efficient online test-time gradient descents but meanwhile cost …
JOIN US
Shape the Future of AI with Sony AI
We want to hear from those of you who have a strong desire
to shape the future of AI.